Nvidia researchers question whether huge context windows are worth it
What happened
The Decoder reported on November 22 that GPT-4 shipped in March 2023 with an 8,192-token context window, while GPT 5.6 Sol now handles about one million tokens, roughly 120 times more. Nvidia's research team argues that feeding a model an entire novel in one pass is often excessive and expensive. Across several tests, periodic summarization plus targeted lookup matched the results of very large contexts, so the team questions whether the push for bigger windows is necessary.
Written by AI from the coverage · updated yesterday
Coverage
Follow the reports to see the story from different sides.
- The DecoderNvidia researchers argue huge AI context windows are a dead end
Nvidia 研究团队认为,模型一次性吞下整本小说这类超大上下文窗口往往是过度且昂贵的做法,在多项测试中,定期摘要加定向查找的效果与之相当。GPT-4 于 2023 年 3 月发布时上下文为 8,192 tokens,如今 GPT 5.6 Sol 可处理约一百万 tokens,约为前者的 120 倍。
Heat over time
Heat now 5·Comparable peak 10(Nov 22 01:00)·Comparable change over 24 hours -49%
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.